With prosperity and development of services such as online education, show anchoring, and live games, a real-time multi-person audio and video call service backend also bears an increasing load. For example, a real-time multi-person audio and video call service based on IM/social tool group membership has a good real-time capability and features delicate network quality control. However, in the service, a network control policy system usually needs to perform a large amount of centralized real-time calculation according to the status of each participate user, and resource consumption increases linearly with increase of a quantity of the participate users. Secondary, to guarantee good call experience, usually there are frequent broadcast notifications (such as in and out of a person skilled in the art, switch of video resource, and up/down microphone) in such the type of service, and usually expansion in the scale of an audio/video room also indicates linear increase of the quantity of the notifications. Finally, full amount/incremental data synchronization usually needs to be maintained between the network control policy system and a data forwarding system, so as to guarantee consistency of user state information in the room in such the type of service. Therefore, a backend performance bottle is extremely apt to appear in a super large scale audio/video room. In particular, when a local fault appears, a centralized broadcast effect in the super large scale audio/video room may further expand the range of affect by the fault, resulting in service instability of the whole system.